| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Mon, 24 Dec 2018 00:29:18 +0530
From: Taniya Das <tdas@...eaurora.org>
To: Matthias Kaehlcke <mka@...omium.org>
Cc: "Rafael J. Wysocki" <rjw@...ysocki.net>,
Viresh Kumar <viresh.kumar@...aro.org>,
linux-kernel@...r.kernel.org, linux-pm@...r.kernel.org,
Stephen Boyd <sboyd@...nel.org>,
Rajendra Nayak <rnayak@...eaurora.org>,
devicetree@...r.kernel.org, robh@...nel.org,
skannan@...eaurora.org, linux-arm-msm@...r.kernel.org,
amit.kucheria@...aro.org, evgreen@...gle.com
Subject: Re: [PATCH v1] cpufreq: qcom: Read voltage LUT and populate OPP
Hello Matthias,
Thanks for your review comments.
On 12/22/2018 2:27 AM, Matthias Kaehlcke wrote:
> Hi Taniya,
>
> On Fri, Dec 21, 2018 at 11:36:48PM +0530, Taniya Das wrote:
>> Add support to read the voltage look up table and populate OPP for all
>> corresponding CPUS.
>>
>> Signed-off-by: Taniya Das <tdas@...eaurora.org>
>> ---
>> drivers/cpufreq/qcom-cpufreq-hw.c | 32 ++++++++++++++++++++++++++++++--
>> 1 file changed, 30 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/cpufreq/qcom-cpufreq-hw.c b/drivers/cpufreq/qcom-cpufreq-hw.c
>> index d83939a..7559b87 100644
>> --- a/drivers/cpufreq/qcom-cpufreq-hw.c
>> +++ b/drivers/cpufreq/qcom-cpufreq-hw.c
>> @@ -10,18 +10,21 @@
>> #include <linux/module.h>
>> #include <linux/of_address.h>
>> #include <linux/of_platform.h>
>> +#include <linux/pm_opp.h>
>> #include <linux/slab.h>
>>
>> #define LUT_MAX_ENTRIES 40U
>> #define LUT_SRC GENMASK(31, 30)
>> #define LUT_L_VAL GENMASK(7, 0)
>> #define LUT_CORE_COUNT GENMASK(18, 16)
>> +#define LUT_VOLT GENMASK(11, 0)
>> #define LUT_ROW_SIZE 32
>> #define CLK_HW_DIV 2
>>
>> /* Register offsets */
>> #define REG_ENABLE 0x0
>> -#define REG_LUT_TABLE 0x110
>> +#define REG_FREQ_LUT_TABLE 0x110
>> +#define REG_VOLT_LUT_TABLE 0x114
>
> The new names suggest that there is a LUT for frequencies and another
> one for voltages. I don't have access to hardware documentation, but
> from the code and offsets in this driver it seems there is a single
> table at offset 0x110, with a 'row' of 32 bytes per OPP. Within this
> row the frequency (and other values) is located at offset 0, the
> voltage at offset 4.
>
> I'd suggest to keep REG_LUT_TABLE, add a define LUT_OFFSET_VOLTAGE/MV
> (or similar) and adjust the math in qcom_cpufreq_hw_read_lut() to use
> REG_LUT_TABLE as base offset.
>
These names are as per HW documentation and the math is kept as per the
documentation for reading the voltage.
>> #define REG_PERF_STATE 0x920
>>
>> static unsigned long cpu_hw_rate, xo_rate;
>> @@ -75,19 +78,26 @@ static int qcom_cpufreq_hw_read_lut(struct device *dev,
>> void __iomem *base)
>> {
>> u32 data, src, lval, i, core_count, prev_cc = 0, prev_freq = 0, freq;
>> + u32 volt;
>> unsigned int max_cores = cpumask_weight(policy->cpus);
>> struct cpufreq_frequency_table *table;
>> + unsigned long cpu_r;
>
> nit: why 'cpu_r' and not just 'cpu'?
>
> (if it is needed at all, see my comment below)
>
>>
>> table = kcalloc(LUT_MAX_ENTRIES + 1, sizeof(*table), GFP_KERNEL);
>> if (!table)
>> return -ENOMEM;
>>
>> for (i = 0; i < LUT_MAX_ENTRIES; i++) {
>> - data = readl_relaxed(base + REG_LUT_TABLE + i * LUT_ROW_SIZE);
>> + data = readl_relaxed(base + REG_FREQ_LUT_TABLE +
>> + i * LUT_ROW_SIZE);
>> src = FIELD_GET(LUT_SRC, data);
>> lval = FIELD_GET(LUT_L_VAL, data);
>> core_count = FIELD_GET(LUT_CORE_COUNT, data);
>>
>> + data = readl_relaxed(base + REG_VOLT_LUT_TABLE +
>> + i * LUT_ROW_SIZE);
>> + volt = FIELD_GET(LUT_VOLT, data) * 1000;
>> +
>> if (src)
>> freq = xo_rate * lval / 1000;
>> else
>> @@ -123,6 +133,10 @@ static int qcom_cpufreq_hw_read_lut(struct device *dev,
>>
>> prev_cc = core_count;
>> prev_freq = freq;
>> +
>> + freq *= 1000;
>> + for_each_cpu(cpu_r, policy->cpus)
>> + dev_pm_opp_add(get_cpu_device(cpu_r), freq, volt);
>
> Are you sure we want to duplicate the OPP entries for all CPUs in the
> cluster? IIUC the frequencies of the cores in a cluster can't be
> changed individually, hence the cores should have a shared table. I
> think dev_pm_opp_get_sharing_cpus() does what you need.
>
> You currently also add OPPs for invalid frequencies. From my SDM845
> device:
>
> cat /sys/devices/system/cpu/cpufreq/policy4/scaling_available_freq
> => 825600 902400 979200 1056000 1209600 1286400 1363200 1459200
> 1536000 1612800 1689600 1766400 1843200 1920000 1996800 2092800
> 2169600 2246400 2323200 2400000 2476800 2553600 2649600
>
> cat /sys/devices/system/cpu/cpufreq/policy4/scaling_boost_frequencies
> 2803200
>
> ls /sys/kernel/debug/opp/cpu4/
> opp:1056000000 opp:1612800000 opp:2092800000 opp:2553600000 opp:825600000
> opp:1209600000 opp:1689600000 opp:2169600000 opp:2649600000 opp:902400000
> opp:1286400000 opp:1766400000 opp:2246400000 opp:2707200000 opp:979200000
> opp:1363200000 opp:1843200000 opp:2323200000 opp:2764800000
> opp:1459200000 opp:1920000000 opp:2400000000 opp:2784000000
> opp:1536000000 opp:1996800000 opp:2476800000 opp:2803200000
>
> There are OPP entries for 2707200000, 2764800000 and 2784000000 Hz,
> however these frequencies appear neither in available_frequencies nor
> boost_frequencies.
>
>> }
>>
Could you help validating with the patch below?
>> table[i].frequency = CPUFREQ_TABLE_END;
>> @@ -159,10 +173,18 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>> struct device *dev = &global_pdev->dev;
>> struct of_phandle_args args;
>> struct device_node *cpu_np;
>> + struct device *cpu_dev;
>> struct resource *res;
>> void __iomem *base;
>> int ret, index;
>>
>> + cpu_dev = get_cpu_device(policy->cpu);
>> + if (!cpu_dev) {
>> + pr_err("%s: failed to get cpu%d device\n", __func__,
>> + policy->cpu);
>> + return -ENODEV;
>> + }
>> +
>> cpu_np = of_cpu_device_node_get(policy->cpu);
>> if (!cpu_np)
>> return -EINVAL;
>> @@ -205,6 +227,12 @@ static int qcom_cpufreq_hw_cpu_init(struct cpufreq_policy *policy)
>> goto error;
>> }
>>
>> + ret = dev_pm_opp_get_opp_count(cpu_dev);
>> + if (ret <= 0) {
>> + dev_err(cpu_dev, "OPP table is not ready\n");
>> + goto error;
>> + }
>> +
>> policy->fast_switch_possible = true;
>>
>> return 0;
>
> I suppose we want to remove the OPPs when the cpufreq driver is
> unloaded, looks like dev_pm_opp_cpumask_remove_table() should do the
> trick.
>
> Cheers
>
> Matthias
>
diff --git a/drivers/cpufreq/qcom-cpufreq-hw.c
b/drivers/cpufreq/qcom-cpufreq-hw.c
index 7559b87..23338b2 100644
--- a/drivers/cpufreq/qcom-cpufreq-hw.c
+++ b/drivers/cpufreq/qcom-cpufreq-hw.c
@@ -81,7 +81,6 @@ static int qcom_cpufreq_hw_read_lut(struct device *dev,
u32 volt;
unsigned int max_cores = cpumask_weight(policy->cpus);
struct cpufreq_frequency_table *table;
- unsigned long cpu_r;
table = kcalloc(LUT_MAX_ENTRIES + 1, sizeof(*table), GFP_KERNEL);
if (!table)
@@ -110,6 +109,8 @@ static int qcom_cpufreq_hw_read_lut(struct device *dev,
table[i].frequency = freq;
dev_dbg(dev, "index=%d freq=%d, core_count
%d\n", i,
freq, core_count);
+ dev_pm_opp_add(get_cpu_device(policy->cpu),
+ freq * 1000, volt);
}
/*
@@ -126,6 +127,8 @@ static int qcom_cpufreq_hw_read_lut(struct device *dev,
if (prev_cc != max_cores) {
prev->frequency = prev_freq;
prev->flags = CPUFREQ_BOOST_FREQ;
+ dev_pm_opp_add(get_cpu_device(policy->cpu),
+ prev_freq * 1000, volt);
}
break;
@@ -133,12 +136,9 @@ static int qcom_cpufreq_hw_read_lut(struct device *dev,
prev_cc = core_count;
prev_freq = freq;
-
- freq *= 1000;
- for_each_cpu(cpu_r, policy->cpus)
- dev_pm_opp_add(get_cpu_device(cpu_r), freq, volt);
}
+ dev_pm_opp_set_sharing_cpus(get_cpu_device(policy->cpu),
policy->cpus);
table[i].frequency = CPUFREQ_TABLE_END;
policy->freq_table = table;
@@ -245,6 +245,7 @@ static int qcom_cpufreq_hw_cpu_exit(struct
cpufreq_policy *policy)
{
void __iomem *base = policy->driver_data - REG_PERF_STATE;
+ dev_pm_opp_cpumask_remove_table(policy->cpus);
kfree(policy->freq_table);
devm_iounmap(&global_pdev->dev, base);
--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a member
of Code Aurora Forum, hosted by The Linux Foundation.
--
Powered by blists - more mailing lists